1; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
2; RUN: llc -march=amdgcn -mcpu=verde -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
3; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck -check-prefix=CI -check-prefix=FUNC %s
4
5@local_memory.local_mem = internal unnamed_addr addrspace(3) global [128 x i32] undef, align 4
6
7; FUNC-LABEL: {{^}}local_memory:
8
9; Check that the LDS size emitted correctly
10; EG: .long 166120
11; EG-NEXT: .long 128
12; SI: .long 47180
13; SI-NEXT: .long 71560
14; CI: .long 47180
15; CI-NEXT: .long 38792
16
17; EG: LDS_WRITE
18; SI-NOT: s_wqm_b64
19; SI: ds_write_b32
20
21; GROUP_BARRIER must be the last instruction in a clause
22; EG: GROUP_BARRIER
23; EG-NEXT: ALU clause
24; SI: s_barrier
25
26; EG: LDS_READ_RET
27; SI: ds_read_b32 {{v[0-9]+}},
28
29define void @local_memory(i32 addrspace(1)* %out) {
30entry:
31  %y.i = call i32 @llvm.r600.read.tidig.x() #0
32  %arrayidx = getelementptr inbounds [128 x i32] addrspace(3)* @local_memory.local_mem, i32 0, i32 %y.i
33  store i32 %y.i, i32 addrspace(3)* %arrayidx, align 4
34  %add = add nsw i32 %y.i, 1
35  %cmp = icmp eq i32 %add, 16
36  %.add = select i1 %cmp, i32 0, i32 %add
37  call void @llvm.AMDGPU.barrier.local()
38  %arrayidx1 = getelementptr inbounds [128 x i32] addrspace(3)* @local_memory.local_mem, i32 0, i32 %.add
39  %0 = load i32 addrspace(3)* %arrayidx1, align 4
40  %arrayidx2 = getelementptr inbounds i32 addrspace(1)* %out, i32 %y.i
41  store i32 %0, i32 addrspace(1)* %arrayidx2, align 4
42  ret void
43}
44
45declare i32 @llvm.r600.read.tidig.x() #0
46declare void @llvm.AMDGPU.barrier.local()
47
48attributes #0 = { readnone }
49